fix(skills): scope the ten loosest descriptions

[UnbreakableMJ]

A coherence audit found two authoring generations in the language-skill descriptions. Eighteen state their scope precisely — concrete triggers, implicit-phrasing examples, and an explicit Do NOT trigger for X (use Y) boundary. Twelve did not, and they were almost exactly the twelve shortest:

generation	rendered chars	boundary clause
precise (18)	671 – 987	yes
loose (12)	254 – 427	no

The gap isn't cosmetic. The description is all a harness sees when deciding whether to load a skill, so one that names no boundary competes for every request its keywords touch.

The sharpest case

elixir ↔ erlang had 0.59 trigger overlap — the highest in the fleet — with neither excluding the other. Both fired on OTP, supervisor, processes, message passing, let it crash, BEAM concurrency: different languages with different APIs (gen_server vs GenServer), indistinguishable on any query naming the runtime rather than the language.

gleam already declined both and warned that generic BEAM advice "gets them wrong" for its typed OTP. Neither reciprocated.

What changed

Ten descriptions rewritten to the pattern the other eighteen use, with triggers drawn from what each skill actually documents rather than invented — Task.async_stream/3 and ExUnit for Elixir, gen_statem/ETS/rebar3 for Erlang, errgroup/pprof for Go, comptime/std.Thread.Pool for Zig, __close and the stylua/selene toolchain for Lua, finalAttrs/nixfmt-rfc-style for Nix, mut/config.nu for Nu, contracts and record merging for Nickel, Loop_Variant/Global/Depends for Ada, parking_lot/spawn_blocking for Rust.

Two deliberately excluded

• microsoft-rust-guidelines is designed to be universal — "MANDATORY for ANY Rust action". A boundary clause would contradict its purpose.
• spacecraft-brand-guidelines belongs to the palette cluster, not the language family. Its scope question is palette authority, handled with that work.

Two needed a plane statement, not a language boundary

• spacecraft-rust-guidelines now says to load microsoft-rust-guidelines first as the mandatory base, and that it stacks conditionally on top, never as the front door — surfacing precedence rules its own body already carried but its description never exposed.
• spacecraft-nu-guidelines now says it governs Nu code you author, while spacecraft-cli-shell owns whether a one-off command is syntactically correct for the shell that runs it. Both legitimately claim Nushell; they differ by plane, and neither said so.

One boundary checked before writing

A nickel/nix clause was considered and verified first — neither skill references the other anywhere in the tree, so the clause states a true language boundary rather than a relationship inferred from nothing.

A regression I caught by re-probing

Dropping message passing from both BEAM descriptions left that phrase matching no skill at all. Narrowing too far is its own failure, so it was restored to both — they now carry mutual exclusions, so triggering-and-self-resolving beats not triggering.

What is deliberately not fixed

The unqualified query "write an OTP supervisor" still reaches all three BEAM skills. That is correct and unavoidable — nothing in that phrase names a language. What changed is that each skill now tells the reader how to resolve it instead of competing in silence:

before:  elixir, erlang, gleam        (no guidance)
after:   elixir, erlang, gleam        (all three carry mutual exclusions)

Verification

rendered lengths: 699 – 936  (none within the 40-char warn margin of the 1000 cap)
elixir excludes erlang/gleam: True / True
erlang excludes elixir/gleam: True / True
AGENTS.md inventories OK (44 root skills, 131 Standard sections)
frontmatter OK (45 skills checked)
0 license-file problem(s). Checked 45 skills.
skill references OK (49 skills in catalogue)
drift sweep → clean

Bundles rebuilt for all ten in the same commit. No skill body changed — descriptions only.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5f12bab7b2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep the prerequisite skill ID unbroken

When YAML renders this folded scalar, the newline between microsoft-rust- and guidelines becomes a space, so agents see the nonexistent identifier microsoft-rust- guidelines. On Rust concurrency requests that rely on this explicit chain, the mandatory base skill cannot be resolved; keep the identifier together on one line or wrap before it.

AGENTS.md reference: AGENTS.md:L478-L480

Useful? React with 👍 / 👎.